The release is one launcher version that opens a single window in place of the launcher: Get Triforce starts the Triforce Setup the release carries, after checking its sha256, and Play NX2navi starts the game that is built on the PC. Nothing of NX2navi is removed. handover/ holds the source of that window as it was published; the drawn pictures, the Setup program and the release files are not kept in the repository (handover/README.md, "In the repository").
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
A review of the DLSS and frame generation paths against NVIDIA's guides (the DLSS programming guide 310.6.0, the
DLSS-FG guide 310.7.0, Streamline 2.14.16's guides and headers).
What follows them: the texture bias (log2(render width / output width) - 1 on top of the game's own, at each change
of the mode), the jitter (Halton, 8 x (output / render)^2 phases and 16 at least, in render pixels within half a
pixel, the scene drawn with it and the motion vectors made without it), the creation flags (HDR, vectors at the
render size, auto-exposure, depth not inverted) and the constants Streamline gets (unjittered matrices, the jitter
in pixels, the vectors' scale of one over the render size).
What did not: vk_dlssg.c holds a change of generation's state for FG_SETTLE_FRAMES frames, so for up to 12 frames
after the depth, the motion vectors or the camera went away (a menu, a fade, a load) generation stayed on and
nx2sl_frame set no constants and tagged nothing. Streamline's guide asks for the constants at every frame and for
null tags where depth or motion vectors are not valid; it logged "Unable to find 'common' constants for frame" and
"Failed to find global tag 'kBufferTypeDepth'" and used the frame before's (in every play log with frame
generation). Such a frame now gets the last frame's constants with reset set and the three tags as null. Not run in
a game yet: nx2navi_dlssg.dll builds and exports its 15 functions.
What the guides leave open: the sign of the jitter offsets against the shift of the picture, which NVIDIA says to
find by trying, and the bias reaches only the tiled, trilinear samplers, not every texture of the scene. For both
there is a test switch now, read once: NX2NAVI_DLAA_JITTER_TEST=<x>,<y> tells the upscaler the jitter times two
factors, and NX2NAVI_DLSS_LOD_BIAS_TEST=<bias> puts a bias in the rule's place. Neither was measured: the runs
that would do it start the game.
The line about the presents a second (the commit before) came at 59 a second under a limit of 60: it comes under 56
now.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The cache counts its short waits in presents (PRESENTS(n): n presents at 60 a second) and scales them by the presents
a second. Above 60 it always did. Below 60 the rate stopped at 60, so the 120 presents a released texture stays out
of the VRAM pressure's reach (STALE_FRAMES) were 4 s at 30 presents a second and 5.7 s at 21. The game lets go of
textures by the second, as the player moves, so what that window holds grows with its length.
A tester's log (RTX 4070 SUPER, 2560x1440, DLSS Quality with frame generation, texture detail Full, 0.10.2-rc4): the
card went to 11,198 MiB against a budget of 10,600 while the frame rate fell from 63 to 21 fps. The released textures
held 2,644 MiB against their bound of 384, and 2,105 MiB of that was released too lately to go (442 MiB when the
pressure began).
On an RTX 5090 with the budget taken as 10,600 MiB and the tester's output settings, a walk through Hyrule Field at a
30 fps limit: the released-too-lately textures were all in their first 120 presents (the new census line below) and
held 383 and 429 MiB on average at the memory reports of two runs, against 292 MiB at 60 fps. With this change they
are in their first 60 presents and hold 209 MiB. Textures made again after a drop: 295 and 523 before, 511 after.
Not shown: the tester's 21 fps and a card that is really over its budget, which this card does not reach; the memory
in use as the driver counts it stayed at 9.8 GiB in the capped runs.
The rate now follows the presents down to 15 a second (vk_tex_release.h trl_rate_sample, trl_rate, trl_presents),
once eight seconds in a row were under 60, and never in a run's first eight seconds; one slow second among fast ones
still shortens nothing. Every wait PRESENTS counts follows it. NX2NAVI_TEX_RATE_LOW=0 stops the rate at 60 as before,
and one line says when the rate crosses 60. Values at 60 presents a second and above are the ones before.
The census of the released textures over their bound gets a second line: what the ones released too lately wait for
(their first presents after the release, content only the GPU made, the gap before a take-back with or without a
lookup still reaching them). It is what showed which wait held the memory.
tests/tex_release_check has the rate's checks and a table of the wait by presents a second; they fail against the
rule before.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The user asked for the release to go out as 0.11.0. No final 0.10.2 was published: only the previews 0.10.2-rc1,
-rc3 and -rc4 carry that name, and the launcher counts 0.11.0 as newer than each of them, so testers are offered it
on either update channel.
VERSION holds 0.11.0 and the notes move to docs/releases/v0.11.0.md, with a line for those on a preview. Every line
that named 0.10.2 as the release in which something changed names 0.11.0 now: comments in the runtime, the two log
lines that give a version (the removed low-end GPU mode in cfg.c, the page generations in vk_memory.c),
nx2navi.ini.example, docs/building.md, the labels of tests/tex_release_check.c and a comment of
tests/audio_params_check.c. AUDIT.md names the release 0.11.0 throughout and records the earlier name and the
previews once, in its status line. Nothing else changes.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The memory report's system line asked for the heap's summary at every report until one was slow. The summary holds
the heap's lock, and every malloc of every thread with it, while it adds the heap up: 54 and 70 ms in two sessions
on an RTX 5090 PC, and the frame of the first report, 30 seconds into each session, took 55 to 82 ms, the longest
frame of play in both. One slow summary switched the question off, so every session paid for it once.
It is now asked for before play only (the boot and the title screen), and at every report with mem_census = 1,
which already accepts a late frame for the walk of the address space. In play the line gives what the runtime knows
it allocated, the same figure from one report to the next. The rule is in runtime/vk_mem_heap.h with
tests/mem_heap_check, which fails against the rule before.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The sixteen settings the fixed look took away are settings again: the tone mapper choice, gamma correction, mid grey, match tone slope, tone slope, hue clip emulation, exposure, highlights, shadows, contrast, saturation, highlight saturation, blowout, flare, hue correction and colour grading. They have their rows in the settings table with the defaults and ranges they had, their rows in the Screen tab under HDR output's three brightness rows (none of them with the game's own HDR), their texts in the four languages, and their lines in the example ini. The tonemap library has the luminance mapper and the grading of the observer-response mapper again, and the support buffer carries five vec4 for it.
A player who changes nothing sees the picture as before. The observer-response mapper leaves each grading step out when its value is 1, and the slider values are divided, not multiplied by a rounded reciprocal, so 50 and 100 % arrive as exactly 1. Evaluated on the CPU in single precision, the shader with every default gave the same bits as the fixed one for 400,000 random scene colours, gains and whites, in HDR and for the SDR picture; the luminance mapper and the game's curve gave the same bits as before they were removed, and the observer-response mapper with its sliders moved stayed within 0.004 % of the pixel of its old result.
Gamma correction now counts with every tone mapper, 2.2 by default, which is what gave the fixed look its true blacks; before the fixed look it ran with the luminance mapper only and the other two showed the dark tones far above black. Its off value brings that lighter picture back for a player who asks for it. The game's own HDR does not take the setting and stays at 2.2.
Lines for these settings that an nx2navi.ini still holds from before the fixed look count again; the example ini and the release notes say so. Nothing of this was seen in the game.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
A texture whose pages show a write is uploaded again unless a hash of its bytes says they are the same. The hash
covered textures up to 64 KiB; for larger ones it was a switch left off until measured in play (tex_same_big). A
run through Hyrule Field measured it (RTX 5090, about three minutes of running): without it 44,000 to 52,000 second
uploads of 33 to 47 GiB of storage in four runs of six, one 1536x1536 texture 29 times within 300 frames with the
same bytes each time, and 66 thousand million texels decoded on the GPU in a run; with it 214 to 592 second uploads,
24 GiB found unchanged, and 20 to 24 thousand million texels decoded, which is the first uploads. The GPU's upload
work fell from 0.41 to 0.14 ms a frame, and the second uploads' 2.9 s of the submitting thread are gone for about
as much hashing.
tex_same_big = 0 hashes only the small ones, as before.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The textures the game lets go of were kept for three minutes and could hold a quarter of the card's budget, up to
8 GiB with room. A run through Hyrule Field shows what that bought and what it cost (RTX 5090, 5120x2160 with DLSS
Performance, about three minutes of running; the game keeps 4.8 GiB of textures live and loads about 190 a second):
the released textures held 5.3 GiB on average and the video memory in use went from 11.1 to 17.9 GiB, while the
game took back 1,494 MiB of them within 10 s of the release, 171 MiB within a minute and 10 MiB later, beside 33,000
textures it loaded anew. A shorter hold alone gives nothing back: with 20 s the released textures fell to 0.5 GiB
and the free space inside the allocator's blocks rose to 4.9 GiB, since a block returns to the driver only when it
is empty. What the released textures hold at their fullest is what the session costs, so the bound is the lever.
The bound is now a thirty-second of the budget, 384 MiB to 1 GiB (vk_tex_release.h trl_cap_mb), still doubled while
the card and the commit have room, and the hold 60 s, 120 for content only the GPU made. The same run with the new
defaults: 13.2 GiB in use on average against 16.7, 13.8 at most against 17.9, 532 MiB of textures made again over
the run, 77 fps against 75. An 8 GiB card keeps 384 MiB where it kept up to 1.9 GiB.
tests/tex_release_check checks the bound by the budget and the sessions it simulates against the rule before.
Compaction (tex_compact) and 16 MiB blocks were tried on the same run and gave back 0.7 and 1.2 GiB; neither is
changed here.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
With Helpers take small jobs on (replay_small, with replay_pipe) the whole scene flashed for single frames under
DLSS, with the interface intact. The pass hooks keep what the last pass that ran them left: the upscaler's light
target is the one of the last scene pass before the output pass. A group's hooks run in its first chunk, and a pass
that chunk began again (after a barrier, or a texture lookup that waited for its turn) ran them a second time. With
the submitting thread already past the group, that came after the hooks of later passes and put the group's own
target back, so the upscaler took an earlier image of the frame for that frame. The log showed it: the light target
changed to another image and back at each flash.
The rule is in runtime/vk_pass_hooks.h, with tests/pass_hooks_check, which fails against the old rule.
Measured by replaying a bug capture in Kakariko (DLSS Performance, 3 workers, replay_pipe = 4, replay_small = 16,
100 s a run): 51 and 52 flash frames as installed, none with replay_small = 0, none with the helpers off, 52 with the
render target reuse off. Two copies of the installed build made the same way, one with this change and one without:
no flash frame in two runs with it, 373 in one run without.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The user asked for it: the mode no longer gains frame rate, and the mesh shader fallback covers the cards it was written for. Its switch in the settings menu's Game tab is gone with the rows under it, and so are their 57 text ids and strings.
low_end_gpu itself does nothing any more. It is still a known key, so an older nx2navi.ini gives no unknown-key warning, and while it is on the log says once that the mode was removed in 0.10.2. The preset that gave nine keys other defaults is gone, and so is the rule that read the mode's keys only while it was on: each of them (mesh_emulation, tex_clobber, foliage, vram_heap_limit, vram_reserve_mb, upload_batch_mb, mirror_stage, mirror_hot, submit_finish_gpu and the others) is an ordinary setting with the default of its row or of its reader, so a line a player wrote by hand keeps counting and a file with only the switch runs the game as it runs for everyone else. guest_opt in nx2navi.ini counts for the next rebuild without the switch too (tools/make_game.py).
The recorder thread read the switch itself: on an AMD card without mesh shaders the mode turned it on. It is off on every AMD card now unless vk_recorder = 1 is set.
nx2navi.ini.example lists the former mode's settings as single settings, the 0.10.2 notes have a Removed entry and name the foliage fix of the commit before this one, and the comments that said what the mode sets say so in the past tense. tests/cfg_parse_check has the new reading of the switch and of the keys without it; the test matrix is at 90 ok with the same two not ok as before (installer_test, worktree_test), and a clean build compiles 169 runtime files and links. Not run in the game.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The low-end GPU mode sets mesh_emulation = 1, and on a card without mesh shaders the emulation then draws the game's task and mesh programs in place of the mesh fallback. On NVIDIA's cards without mesh shaders that ended the game at the first mesh draws of an area with an access violation inside the driver, at nvoglv64.dll+0x10cdffe on a GTX 1050 Ti and on a GTX 1070 (both 0.10.1), and a Titan X (Pascal) report names the same mode. The GTX 1050 Ti's owner set mesh_emulation = 0 and played on through the fallback with the rest of the mode on.
The decision (vk_mesh_path.h) takes the card's vendor: an NVIDIA card is drawn for as if the emulation were not asked for, and the boot line says so. The replay and the pipeline code read the decided path from the device context where each read the setting itself. tests/mesh_path_check.c has the new cases.
Not run in the game: no GTX 10 card here. mesh_path_check passes and every file that includes the changed headers compiles.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Four players of 0.10.1 could not start the game again after changing settings: "no usable Vulkan device", exit code 70, until nx2navi.ini was edited or removed by hand. One log shows the case: a laptop with an RTX 3050 and Intel Iris Xe graphics, framegen = dlss and gpu_index = 1, the Intel one. With frame generation on, the device is made through Streamline's interposer, which asks it for VK_NV_low_latency2 and failed with VK_ERROR_EXTENSION_NOT_PRESENT, and nothing else was tried.
A chosen card of another vendor now counts as not usable while the interposer is in use, so the best usable card is taken, as for a chosen card that is gone. Where the card in use is still not an NVIDIA one, or the interposer fails to make the device for another reason, the loader makes it and frame generation is off for the run: the log says why and the settings menu shows it as unavailable, as it does when Streamline's files are missing.
Not run: no PC here has two cards of different vendors. The changed files compile.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
0.10.2 was to turn replay_pipe = 4 and replay_small = 16 on by default. A tester measured those two with five replay workers against the defaults on an RTX 3090, standing at eight places in Hyrule Field and at Lake Hylia with the default run repeated (within 3 %): slower at four of them (50.1 instead of 58.1 fps at one) and level or at most 7 % faster at the other four. The user decided to leave both at 0. CPU helper threads stay Automatic and Start next frame early stays on, as 0.10.1 shipped them; the notes and nx2navi.ini.example say so.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
auto chose full from 8,192 MiB of video memory budget, which is the cards of 10 GiB, and went by the budget alone. Two reports on 0.10.1 say that is too much. A tester filled a 24 GiB card to the room of an 8 GiB one and toured eleven areas at 2560x1440: with full, the driver counted more than the card had (the textures and the free space the allocator's blocks keep, which the pressure check leaves out), memory went to system memory, and the tour ran at 29.6 fps with a 1% low of 11.4; with reduced, 42.8 and 36.4, and nothing spilled. And a handheld's integrated GPU reported a budget of 10,361 MiB, which is a share of its 16 GiB of system memory: auto chose full and the game ended when a save was loaded, until its player set reduced.
Full now starts at 10,240 MiB of budget, the cards of 12 GiB (one reported 11,313 MiB in a player's log), and an integrated GPU gets reduced at most by auto. The log line says so when it applies, the settings menu's automatic entry names the level as before, and a level the player set stays.
Not run in the game. The file compiles and the content check passes.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The game draws a culled list of grass or leaves twice from one indirect record: into the depth target alone, then into the colour targets. On the mesh shader fallback each draw has compute modules and a pipeline of its own, made at its first draw, and the depth-only one is ready first. Drawn alone it left the list's depth in the scene with nothing drawn over it, so what lies behind failed the depth test there and the sky's colour showed through the ground in the shape of the grass for as long as the colour draw's pipeline built, which is seconds each on AMD's driver for Radeon RX 6000 and older, where the fallback is the default.
A colour draw skipped for its modules or pipeline now stamps its record's address with the present, and clears the stamp when it is drawn. The depth-only draw of a record stamped within the last two presents is skipped as well, counted like a draw whose own pipeline is building. A draw whose stages write memory is never held. mesh_fallback_hold = 0 draws the depth alone as before.
Not run: no card here takes the fallback by default, and the game is the user's to start. The changed file compiles and the settings table's check reports 0 problems.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Main gained three commits after 0.10.1's release commit: a render target that was drawn over gives its image to the next target of its size, such an image is handed on after 8 presents and freed after 32 while the card is past its pressure share, and 0.10.1 kept CPU helper threads on Automatic and Start next frame early on. The settings keep this line's defaults (Helpers work ahead 4, Helpers take small jobs 16), the comments that said both were off in 0.10.1 now say what 0.10.1 shipped, the texture log keeps both sides' counters, and 0.10.2's notes say which two defaults are new.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The review found that dynres.c dynres_frame compared a changed target with a static that started at 0 and outlived a
restart: with fps_limit = display the first refresh-rate wander under 1 % after the start was logged, against the
comment's rule, and after dyn_res off and on a move back to the earlier run's last logged target said nothing, so the
start line and the live lines disagreed. A test of the menu's path found a third case: a limit or dyn_res_target
changed in the menu bumps cfg_version, dynres_read takes the new target first, and dynres_frame, which logged only
when its own reading differed from D.target_ns, never named it. The last logged target now lives in D (logged_ns),
the start line sets it, and dynres_frame logs any move of more than 1 % from it, naming the set target as "as set".
No lock and no wait; the target and the steps are what they were. The review's stale "(dyn_res_target 16.7 in
1..100)" in cfg.c misfit now reads 0..100, as the row's set has been since the commit that changed the default
target.
tests/dynres_check.py gains these cases (a menu limit, a menu target, a wander after the start, the earlier run's
target after off and on); they fail on the old dynres.c and pass. The -Wall compile of dynres.c and cfg.c shows no
warning, and cfg_rows_check reports 0 problems.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The review found that the header promised more than cfg.c does in two places. It said a blank then '#' or ';' after
any number starts a comment, but cfg.c cut_comment cuts only for a row with a set of values, so the settings that
take any number (gpu_index, compute_keep, tex_released_max, floor_probe, floor_scan) keep the comment in the value:
"compute_keep = 256 # less" reads the default 512. And it said a setting of 0 or 1 takes off, false and no, but
switch_word maps the words only for a row whose set is exactly 0|1; a setting with choices 0, 1 or 2 such as
async_preempt or tex_disuse reads "off" as its default, on, with no line in the log. The header now says the comment
is cut after the value of a setting with a list of choices or a range, names those five number settings among what
keeps its whole text, and gives the words to the settings whose only values are 0 and 1. docs/building.md's paragraph
on tests/cfg_parse_check.py now also names the low-end switch's case and make_game.py's tests added by the previous
commit. Comments and docs only; cfg_rows_check reports 0 problems.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A hand-edited "pace_smooth = 0 ; off" kept the comment in the value, which the readers took as on, "fps_limit = 144 #
monitor" gave 60, and "fullscreen = false" or "rtss = off" were on, since every reader took anything but "0" as on;
the log only named these lines (the check added for F-7). Now, for a key whose row has a set of values, cfg_load cuts
an unquoted value at the first blank followed by ';' or '#' after the '=' ("key = ; note" is unset) and logs the line
once; free text keeps its whole value (window_size, monitor, audio_device and keys without a row such as mods_dir,
pipeline_cache and the bindings, where a path or name may hold " #"), as do a value in double quotes and the
environment's values. A switch (a row whose set is 0|1) written as false, off or no reads as 0 and as true, on or yes
as 1 through cfg_get itself, so every reader agrees (cfg_bool, rt_knob, the readers that compare with "0" or parse a
number); cfg_bool and rt_knob also take those words for keys of other rows, and rt_knob no longer logs them. F-7's
check now names a switch value only when it is none of those, and a comment kept in a value only where it stays
(free text, the environment). The launcher's reader (launcher/gameini.py) cuts the same way for keys the caller
says have a set of values; its mods keys stay free text. The menu's save still writes a changed key's line again
without its comment.
This changes what existing files with such lines do, which is what was approved. tests/cfg_parse_check.py,
tests/rt_knob_check.py and launcher_test PlayerIniTest cover the cut, the words, free text, quotes and the
environment, and pass; cfg_rows_check reports 0 problems; the runtime compile check and the -Wall comparison of cfg.c
and perf.c show no new warnings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Dynamic resolution's target was 16.7 ms unless set, whatever the frame limit, so under a 120 or 144 fps limit the GPU
could take up to 18 ms a frame at full size before a step down, and the game drew at about 55 fps instead of the
limit. Each step was held for 45 presented frames, 0.75 s at 60 fps but 0.19 s at 240, so the size changed four times
as often there. The target now defaults to 1000 / the rate the game draws at under the limit (frame_rate.c
frame_step_game_hz: the monitor's rate for display, the drawn frames with frame generation), read every frame so a
limit changed in the menu moves it, and 16.7 ms while that rate is unknown or unlimited; dyn_res_target set to 1 or
more still wins (its row's default is now 0). dyn_res_step keeps its key, range and default, now counted in 60ths of
a second of the clock, so 45 is 0.75 s at any frame rate and the same as before at 60 fps. The switch stays off by
default.
Not reproduced before the change (read from the code). The new tests/dynres_check.py builds dynres.c with stand-ins
and passes (24 failures on the old code); cfg_rows_check reports 0 problems; the runtime compile check and the -Wall
comparison of dynres.c show no new warnings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
shake_stick picked the side from the time since the press, 67 ms a side (two frames at the game's 30 fps), so above
30 fps each side lasted more of the game's updates: 8 at 120 fps and 16 at 240, slower swings than the game reads at
its own rate. The side now comes from the game's present count, which frame_rate.c already kept for the flight
recorder and now gives through frame_step_presents, two presents a side. The press still swings for 800 ms of the
clock and a hold keeps it going, as before; while no present comes (a load) the stick keeps its side.
Not reproduced before the change (read from the code). The new tests/shake_key_check.py builds shake_stick from
input.c and passes (it fails 118 times on the old code); the runtime compile check and the -Wall comparison of
input.c and frame_rate.c show no new warnings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The ocarina manager counts the length of each note and rest in whole 60 Hz frames and adds the time manager's frames
per update to that clock at four sites (main+0x32926c8, main+0x3293410, main+0x3296b08, main+0x3296da4) through
fcvtzu, which cuts the step toward zero. Above 60 fps the step is under one frame, so the clock never moved: a
recording kept its notes without their lengths (the recorder leaves a length of 0 out), the 360 and 120 frame limits
never came, and a recording ended after a fixed 60 frames of its timer instead of following the clock; at the 40 and
45 fps limits it counted one a frame. The set replaces each fcvtzu with nine words that add the fraction carried on
the patch page's +264, store what is left after the cut (NaN and negatives as 0) and convert the sum. Steps of one
and two frames give the game's count with the fraction left at 0, so 30 and 60 fps are unchanged. The words call
nothing and take no lock (one of the functions returns through an unsaved x30), and use only v16-v18, which the three
functions never name. The set is appended so that no existing set's index moves; ocarina_timing = 0 gives the game's
words.
Not reproduced before the change (read from the code). The generator's check of the original words passed against
the 1.1.1 main, the new tests/ocarina_timing_check.py passed (.venv, Unicorn), as did frame_step_check,
text_speed_check, sprint_check, movie_timeline_check, cfg_parse_check, launcher_test and cfg_rows_check.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The review found that audio_mixer.c splitters_walk used AUDIO_REFUSE for a framing fault in both of its walks. In the
apply walk (audio_mixer_update_routing, after audio_mixer_routing_check's walk passed) such a fault can only come from
the game rewriting its input during its own RequestUpdate. The update is taken by then: it is answered 0, its voices
and effects and the splitter records before the fault are applied, and the mix section is applied after it. Yet the
log said "RequestUpdate refused with 0x5299 ...; the renderer goes on with the last update it took". SPL_FAULT now
refuses the update in the check walk as before, and in the apply walk logs the fault with "the rest of the splitter
section is left out". The same double read of the mix section by audio_mixer_mix_layout_check and
audio_mixer_update_routing stays as it is; its comment now says the check holds for the apply only because the game
does not rewrite its input during its own RequestUpdate. docs/building.md's audio_params_check paragraph names the new
case and is wrapped again at 120 columns.
Not reproduced before the change: the game's sending thread waits in the IPC call, so neither case has a known way to
happen. A new case in tests/audio_params_check calls audio_mixer_update_routing with a splitter section whose first
destination record's magic is broken after a valid update. It checks that the splitter info before the fault and the
mix section are applied, that the destination record is not, and that the one log line names the rest of the section
left out. It failed on the code before this change ("RequestUpdate refused with 0x5299: splitter destination 0 ...
not SNDD"). audio_params_check (27 refused updates, 11 left-out records, 38 effects and sinks, both strict runs),
check_content, a compile of every runtime file and -Wall -Wextra on the audio files and cfg.c with no warnings passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The review found six texts that claimed more than the code does. NX2NAVI_AUDIO_STRICT=1 does not bring 0.10.1 back
exactly: it also ends the process on a limiter attack or release that is not a number and on a mix in use with a
negative buffer count, which 0.10.1 let through, so the cfg.c comment, nx2navi.ini.example, the audren.h audio_bad
comment and the test's comment now say it does so for most of these values. Its FATAL line described the non-strict
outcome ("its sound passes through unchanged", "the renderer goes on") and then ended the process; it now ends with
"NX2NAVI_AUDIO_STRICT=1 ends the process here instead". The audren.c header and docs/architecture.md said nothing the
game sends ends the process, while OpenAudioRenderer's parameters and the request's buffer descriptors still do; both
now say a value in the game's RequestUpdate does not, and name the two that still do. The first log line of a kind
gave "N such values so far", but bad_count counts every update and every render pass that meets any kind, so it now
reads "N times so far for all kinds together"; the renderer_close line and the bad_count comment say the same.
docs/building.md and the test's comment called 0x5299 the one result the SDK hands back; the SDK hands back 0x5299,
0x5499 and 0x5699 and aborts on any other. Texts only: no behaviour changes.
The first commit's message says any other result than 0x5299 makes the SDK abort; it should read any result but
0x5299, 0x5499 and 0x5699. The branch's history is kept, so this message carries the correction.
audio_params_check (the strict runs still end with "FATAL: audio:" and the switch's name), cfg_rows_check (0
problems) and check_content passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The audio renderer checked the game's RequestUpdate with about 80 rt_fatal and need() sites in audren.c,
audio_mixer.c, audio_effects.c and audio_reverb.c, so a section size, a splitter or mix record or an effect setting
out of range (a reverb3d whose delays add up past 0.4 s, a limiter attack outside 0 to 1, a mix buffer index past its
mix, a filter coefficient that is not finite) ended the process. The tested route never reached one; dungeons and
bosses were never run. Now an update whose sections do not match the renderer is refused whole with 0x5299 before
anything of it is applied: the SDK's RequestUpdateAudioRenderer hands that result back to the game, which ignores it
and sends its next update, where any other result makes the SDK abort. A record with an id the renderer does not have
is left out of an update that is otherwise applied, a splitter destination id it does not have is stored as -1, which
both render loops skip, and an effect or sink whose values cannot be used is passed through or skipped in the render
pass, with every index checked before its first write. audio_bad logs the first of each kind per renderer with the
game frame and counts the rest. rt_fatal stays for host allocations, OpenAudioRenderer's parameters and the checks
the new ones turn into invariants. Two inputs that did not end the process before change too: a mix in use with a
negative buffer count is refused (it laid the mixes after it out before buffer 0), and a limiter attack or release
that is not a number passes the sound through instead of turning the mix into NaN. NX2NAVI_AUDIO_STRICT=1
(audio_strict, default 0) makes audio_bad end the process as before, for bug reports.
Not reproduced before the change in the game; crafted updates reproduced the fatals. The new
tests/audio_params_check sends 27 refused updates, 10 records and 38 effects and sinks the renderer cannot use
through renderer_handler and runs the strict switch; 15 deliberate breakages of the change each made it fail.
audio_mixer_check, cfg_rows_check, check_content, a compile of every runtime file with no new -Wall -Wextra warnings
in the changed ones and the stability pass's test matrix (the two baseline environment failures only) passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The review of 6a0f123a found that nvn_submit_prepare took the same handle given twice in one nvnQueueSubmitCommands
call for another thread's submission in flight: the first occurrence takes the copy's busy count from 0 to 1 and
resolves in place, so nvn_copy_enter returns 0 for the second, and unless the copy was known to hold no deferred binds
(deferred 1; it is 0 at the first submission and 2 for any copy with deferred binds) the call cloned it, a malloc and
copy of up to 1-2 MiB and a pin walk under copy_lock shared. It also logged "thread N submits the recording ... while
another thread's submission of it is replayed", the line the play-test reads to tell whether F-2-03's remainder
happens in this game at all, so a call that repeats a handle would have given a false positive. The asynchronous path
already keeps such a repeat in place (nvn_submit_prepare_async: used_at != now).
Now runtime/nvn_submit_guard.h nvn_copy_again says whether the handle came earlier in the same call and resolves in
place there; nvn_submit_prepare asks it only after nvn_copy_enter found the copy busy and the copy may hold deferred
binds, so only that branch pays for the scan. Such an occurrence stays in place and keeps its busy count, which
nvn_submit_done leaves per occurrence as before. Another thread that submits the copy meanwhile still takes a clone.
tests/submit_clone_check gives the copy twice in a quarter of its calls (76,759 repeats kept in place in 800,000 calls,
no replay saw its handles change, the counts end at 0) and checks nvn_copy_again step by step: given twice with
nothing else in flight, in place both times; with another submission in flight, cloned both times.
docs/building.md says so.
Not reproduced before the change. Checks that passed: submit_clone_check, pass_cc_check of nvn_api_cmd.c and
nvn_api_core.c with no new -Wall -Wextra warnings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The review of cd2fd454 found that with present_order = 1 the present thread could record and queue frame N's blit
before frame_end(N) ran: when the presenting thread's try of the submission lock failed, frame_end(N) waited for the
next submission, and that could be the present thread's own blit submission, which present_main makes only after
vk_present_blit. The blit reads what vk_dlaa_frame resets when the upscaler or frame generation is switched on or off
(gfx_present.c hudless_record through vk_dlaa_fg_hudless; vk_fg.cpp, vk_fg4.cpp and vk_dlssg.c through
vk_dlaa_fg_inputs, vk_dlaa_fg_views and vk_dlaa_fg_camera). vk_dlaa_frame drops those inputs at a switch because
handing them to frame generation ended an AMD process before, and in 0.10.1 vk_dlaa_frame(N) always ran before blit
N. So a switch frame inside an overlap window could hand frame generation the inputs the switch should have dropped.
Now present_image tells the present thread when it left the frame owed (PresentJob.owed), and present_main first
makes sure that frame's bookkeeping ran (present_frame_closed, vk_frame_owed.h fo_close): when a submission did it
meanwhile, done says so (it is now stored with release after each frame's bookkeeping and read with acquire, fo_ran)
and no lock is taken; otherwise the present thread takes the submission lock, waiting for a replay in progress
(vk_sync.c vk_sync_locked), and runs what is owed. frame_end(N) therefore always comes before blit N, as in 0.10.1.
The present thread is not one of the game's threads, and it waits for the same lock for the blit's submission a
moment later anyway. It waited for that lock with its busy flag set before too, and nothing that holds the lock waits
for the present thread, so no deadlock is added. The cost, only for a frame still owed when the
present thread takes it: the blit's recording no longer overlaps the replay it waits for, which can lengthen a drawing
thread's existing wait in vk_present_drain by that recording. The common frame, whose present got the lock, takes no
lock here. The log's frame bookkeeping line (exit, and trace = perf every 600 presents) counts these frames and the
time they took the present thread, the wait included.
tests/frame_owed_check now has a present thread that takes each frame from the presenting thread (one in waiting at
most), blits it and submits: no blit comes before its frame's bookkeeping. Its "old" argument leaves the catch-up out
and fails, as the branch had it: 2608 and 2706 blits before their frame's bookkeeping in two runs of 20,000 frames.
docs/building.md, docs/architecture.md and nx2navi.ini.example say where the bookkeeping runs.
Not reproduced before the change. Checks that passed: frame_owed_check (20,000 frames; 12,695 owed, 1,888 of them
caught up by the present thread, 0 blits before their bookkeeping, the longest present 0.003 ms), pass_cc_check of
every runtime file (headers changed), no new -Wall -Wextra warnings in vk_backend.c and vk_sync.c.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The submit path no longer relies on one guest thread at a time: replays take turns on the submission lock, a frame's
bookkeeping at its present runs between two replays (vk_frame_owed.h), a held present goes out once and in order
(vk_present_held.h), a second submission of a recording in flight gets a clone (nvn_submit_guard.h), and held work is
asked about from a published copy (vk_evo_peek.h). docs/architecture.md section 4 gets a bullet saying so, with the
switch and the log lines; the comment above nvn_api_core.c nvn_call_note now says what holds and that an overlap line
is a count, not a fault.
Checks that passed: pass_cc_check of nvn_api_core.c.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
nvnQueueSubmitCommands resolves a copy's deferred binds at the submit call: the handles the game's arrays hold now are
written into the recording copy (nvn_api_cmd.c resolve_deferred). With the replay on the submitting thread, a second
guest thread that submitted the same copy while the first one replayed it, or while the first copied it for a held
continuation (vk_exec.c cont_new), rewrote those handles under that replay: the rest of it bound the next frame's
textures and samplers, and a held continuation could be copied torn (F-2-03's part left after the private copies).
Now each copy counts the synchronous submissions between their prepare and the end of their replay (copy_prefix.busy,
which shares the 8 bytes pins had, so the prefix keeps its 56 bytes). nvn_submit_prepare takes copy_lock shared once
for all handles: the first submission in flight resolves the copy in place, as before; one that finds another in
flight, unless the copy is known to hold no deferred binds, resolves a clone made under that lock, with its texture
pins counted, and nvn_submit_done frees it after the replay, as the asynchronous path does for a copy a queued
submission still reads. A copy the game releases while a submission of it is in flight becomes a zombie and is freed at
the next sweep after it ends (release_copies, zombie_sweep). The bind loop moved unchanged into
runtime/nvn_submit_guard.h (nvn_resolve_binds) so a check can build it; it reports a corrupt record or a call past 16
deep to resolve_deferred, which ends the run with the messages it had, and keeps the first six changes' trace line.
Being code the front end carries under its MIT header, the new header keeps that line and THIRD_PARTY_NOTICES.md's row
names it. A clone is logged the first four times and at each power of two ("submits the recording ... while another
thread's submission of it is replayed"). No timing change and no wait: the clone path runs only when two threads have
one copy in flight at once.
Not reproduced before the change; nobody knows yet whether the game ever submits one copy from two threads at once (the
new line will say). Checks that passed: the new tests/submit_clone_check (four threads submitting two copies 200,000
times each while their arrays change: no replay saw its handles change, both paths ran, the counts ended at 0; the
unguarded 0.10.1 shape through the same harness: about 450,000 of 800,000 replays saw them change; nvn_resolve_binds
step by step), copy_sweep_check (its stand-in prefix follows the new layout), pass_cc_check on every runtime file, no new
-Wall -Wextra warnings in nvn_api_cmd.c and nvn_api_core.c.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
present_image ran the frame's bookkeeping on the presenting thread without a lock: vk_exec_frame_log (the trace counts
and vk_exec_cond_frame_end's frame count, which the conditional draws go by), dynres_frame, vk_exec_barrier_log,
vk_dlaa_frame (the DLSS mode and its LOD bias, the frame's points, the swap image, the idle releases), vk_mvsurvey_frame,
vk_mvprev_frame (which swaps the two motion tables the draws insert into) and vk_gpu_passes_frame. All of it is state the
replay reads and writes as it records, under sub_lock. With the replay on the submitting threads the game presents on
one thread while another may be replaying (the overlap lines of the 0.10.0 fresh session, at area changes), so the
present reset it in the middle of that replay: a DLSS frame reset halfway, the motion tables swapped halfway (with
MVPREV_CHECK=1 an index out of bounds), the conditional draws counted against the next frame.
Now the bookkeeping is frame_end, and runtime/vk_frame_owed.h decides where it runs. present_image marks its frame owed
and runs it itself only when it gets sub_lock at once (vk_sync.c vk_sync_try_locked); otherwise vk_sync.c submit runs
every owed frame's, in frame order, right after it takes the lock and before its replay (vk_present_frames_owed), on
whichever thread submits next, the present thread's blit submission included. The presenting thread never waits for
another thread's replay for it: no new wait on a drawing thread; the thread that takes the lock next does the work it
would have done at the present, tens of microseconds normally. The hand-off, vk_texture_get with its upload flush and
vk_present_drain stay outside the lock, as frame_end must not wait for the present thread or submit; its callees were
checked for that. vk_sampler_aniso_live and vk_mem_census keep their place in present_image (their own locks). In the
asynchronous replay the other holder of sub_lock is the present thread's blit, and the same applies.
NX2NAVI_PRESENT_ORDER (present_order in [diagnostics], default 1, read at start) puts it back at the present with 0, as
in 0.10.1: metas, known and sections rows and an nx2navi.ini.example entry.
Not reproduced before the change. Checks that passed: the new tests/frame_owed_check (one presenting and three
replaying threads, some replays holding the lock 40 ms, over 20,000 frames: every frame's bookkeeping once, in order,
never during a replay and before any replay that took the lock after its present; the longest present under 1 ms with
19,000 presents made during a long hold), cfg_rows_check with 0 problems, pass_cc_check on every runtime file (headers
changed), no new -Wall -Wextra warnings in vk_backend.c, vk_sync.c, vk_exec.c and cfg.c.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With the replay on the submitting thread, a present that waits for event-ordered work (vk_backend.c PH) goes out after
the submission that replays the last of that work, at the next present, or after PH_MAX submissions. The game submits
from several threads while one of them presents, so that submission may be another thread's. PH had no lock: two
submitting threads could both pass ph_flush's check of PH.on and put one frame out twice; a thread reading PH's fields
while the presenting thread armed the next frame could put out a job mixed from two frames; and a held frame put out on
a submitting thread could reach the present thread after the presenting thread's next frame, a picture back in time.
runtime/vk_present_held.h now holds PH's job, its arming, putting out and counts behind PH.lock. Submitting threads
only try the lock (ph_flush_try) and go on when another thread holds it: that thread is putting a present out or arming
one, and the next submission, the next present or PH_MAX checks again, the bound there was before. The presenting
thread takes it for every synchronous present while held presents are on (ph_flush_next, ph_present), so the
synchronous present_image calls run one at a time and in the game's order. That is the one new wait, and only the
presenting thread makes it: only while another thread puts a held present out, for at most that present_image. It is
counted (lock_waits, lock_wait_ns, lock_wait_max) and timed as PERF_DRAIN. No path holds the lock across held_resume,
which is a replay. Neither 0.10.0 fresh log has a held present at all ("went out" and "presents that waited" lines),
so the wait is not expected in normal play. present_held = 0 turns held presents off, and with them every use of the
lock.
Not reproduced before the change. Checks that passed: the new tests/present_held_check (one presenting and three
submitting threads over 200,000 frames: every frame out once, in order and never two at once; the unlocked 0.10.1
shape through the same harness gave about 205,000 presents for 200,000 frames, 6,000 to 8,000 out of order or twice and
8,000 to 12,000 at the same time; step-by-step cases; a submitting thread gives up on a held lock at once and the
presenting thread's wait is counted), pass_cc_check on every runtime file, no new -Wall -Wextra warnings in
vk_backend.c.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The review of 9071dfc3 found that nx2navi.ini.example, cfg.c's metas comment, the commit message and docs/architecture.md
said the rollback switch works "as up to 0.10.1", which it does not exactly. With 1, every place that lets something go
waits for the device again, C-5's queue-lock hazard with it, but what was let go of still goes through vk_dlaa.c's list
and vk_memory.c's deferred destruction, the motion vectors and the HUD-less picture keep their 8-present gate,
vk_upscaler_settle leaves what the submission being recorded names, stale_sizes waits once per call, and a refused DLSS
creation with features let go of still listed waits and tries again, where 0.10.1 turned DLAA off. Each difference is
the safer side. The ini line now says that the game waits at each change as up to 0.10.1 while what was let go of
still goes once the card is past it; cfg.c, vk_dlaa.c's "what the upscalers let go of" and docs/architecture.md say
the same. Comments and docs only.
Checks that passed: cfg_rows_check (0 problems), upscaler_idle_check, cfg.c and vk_dlaa.c compile with -Wall -Wextra
and no warning, tools/check_content.py --staged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The review of 2b9b3061 found that the check missed the dangerous regression. It only checked that each place that lets
something go calls vk_upscaler_let_go, so with destroy_size put back in vk_dlaa.c free_size, or hl_destroy in
hudless_from's size change, it still passed, while either frees images or a DLSS feature a frame still in flight uses.
The check now also fails when:
- a function that lets go of what the GPU may still use (vk_dlaa.c free_size, free_snap, stale_sizes, release_kept,
retire_idle, evaluate and the retire_size, retire_feat, retire_snap, retire_hl and retire_pb helpers; vk_ffx.cpp
context_for, trim, retire_up, vk_ffx_trim_idle and vk_ffx_release_idle; vk_ffx4.cpp context_for, vk_ffx4_trim,
retire_up and vk_ffx4_release_idle) calls a destroy that frees at once (destroy_size, destroy_snap, hl_destroy,
pb_destroy, feat_release, N.release, vkDestroyImageView, vmaDestroyImage; A.destroy in the FSR files);
- hudless_from, pb_images or snap_for call one of those before their first image2d call (after it, only their half-made
new images are destroyed);
- upd_take_all, which releases a whole list whatever its gates, is called outside teardown (vk_dlaa_release,
vk_ffx_release, vk_ffx4_release) and a full list's fallback (dead_put, the FSR files' retire_up), or there before
vk_upscaler_settle's wait for the device.
docs/building.md "Checks" says so.
Each of 19 mutations of scratch copies fails the check (destroy_size in free_size and retire_idle, hl_destroy in
hudless_from, pb_destroy in pb_images, destroy_snap in free_snap and snap_for, feat_release in release_kept, N.release in
stale_sizes, vmaDestroyImage in retire_size, upd_take_all in retire_hl, A.destroy in FSR 3.1's retire_up and in either
file's trim and release_idle, dead_put and FSR 4's retire_up without the settle, FSR 3.1 and FSR 4 without the retry);
the unchanged
sources pass, and stability-pass's sources fail with 46 problems. tools/check_content.py --staged passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The review of 09027478 found that FSR 3.1 lost its fallback for a refused creation. Up to 0.10.1, vk_ffx.cpp
context_for waited for the device and destroyed the other render size's or the least recently used context before
create_counted, so their video memory was free for the new one. Since 09027478 they only go into g_dead, and the drain
in vk_upscaler_let_go destroys those whose last upscale the GPU is past. When it is still behind and the library
refuses the new context for want of memory, context_for returned nullptr, vk_ffx_upscale 0, and evaluate's dl_fail
turned DLAA off for the run. DLSS (evaluate) and FSR 4 (context_for) already wait for the GPU to pass what they let go
of and try again.
context_for now does the same when the creation is refused and FSR 3.1's or FSR 4's list of contexts let go of is not
empty: vk_upscaler_gpu_wait (a timeline wait for those submissions, not the device's idle; the device with
upscaler_idle_wait = 1), vk_ffx_dead_take, and a second create_counted when something was destroyed. A line says the
refusal, the wait and the outcome. upscaler_idle_check now also fails when vk_dlaa.c evaluate or vk_ffx.cpp or
vk_ffx4.cpp context_for has no vk_upscaler_gpu_wait (a scratch copy without the retry fails it), and docs/building.md
says so.
Not reproduced before the change (found by review). Checks that passed: vk_ffx.cpp compiles with -Wall -Wextra and no
warning, upscaler_idle_check, tools/check_content.py --staged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tests/upscaler_idle_check.py reads vk_dlaa.c, vk_ffx.cpp and vk_ffx4.cpp with their comments and strings blanked and
fails when vkDeviceWaitIdle is called outside vk_upscaler_device_idle and the --probe-ffx-cache path, when
vk_upscaler_device_idle is called outside vk_upscaler_settle and vk_upscaler_gpu_wait, or when one of the ten places
that let something go does not call vk_upscaler_let_go, through which NX2NAVI_UPSCALER_IDLE_WAIT=1 brings the old wait
back. It also checks that switch's sections[] entry and nx2navi.ini.example line (cfg_rows_check has its known[] name
and metas[] row), that perf.c names every PERF_M_* kind (a kind without a name prints a null string on a slow frame's
line), and that pb_images notes PB.used after the last call that clears PB.
On this branch it passes. On the base of this branch's work (stability-pass, 1f547853) it reports the twelve waits of
C-6 and Cv-1, each with its function, and on the first commit of this branch, where the waits went through
vk_upscaler_device_idle, the same twelve places. docs/building.md "Checks" describes it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
vk_ffx.cpp context_for (two places) and vk_ffx4.cpp context_for (three) waited for the whole device to be idle, with
the queue lock held, on the thread that replays the game's submissions, before they destroyed an upscale context of
another render size, the least recently used one for a new output size, or (FSR 4) every idle one after a refused
creation: a stall of the present's flip, the completion thread and upload submits at every pause-screen size, reload,
window size, dynamic resolution step or mode change with FSR on.
A context that goes is now let go of (retire_up): it goes into a list (vk_up_defer.h) with the timeline value of the
submission of its last upscale (Up.fence, set after each dispatch) and is destroyed on the same thread once the GPU is
past it, at the points where vk_dlaa.c takes its own list (every evaluation, every pass while none runs, and after
something is let go of). The 90-second trims, vk_ffx_trim_idle and vk_ffx_release_idle (the upscaler off, which no
longer needs retire_idle's wait) let go of theirs the same way. FSR 4's create descriptors, which the API asks to stay
live until the context is destroyed, move from the slot into a calloc'd Made that goes with the context, so the slot
can take a new context while the old one waits. A refused FSR 4 creation now waits on the timeline from the host for
the contexts let go of, no queue touched (vk_upscaler_gpu_wait), and tries again when any went; with
NX2NAVI_UPSCALER_IDLE_WAIT=1 that wait and every let go wait for the device as before. docs/architecture.md says
where the upscalers' releases go.
C-5 is narrowed: on the default path no upscaler waits for the device any more, so these waits no longer touch the
frame generation swapchain's own queues; vk_fg.cpp, vk_fg4.cpp, gfx_present.c, vk_hdr.c, vk_fsr.c and vk_color.c keep
theirs.
Not reproduced before the change (no FSR session log here). Every runtime file compiles, vk_dlaa.c, vk_ffx.cpp and
vk_ffx4.cpp with -Wall -Wextra and no warning; vk_ffx4.cpp's non-Windows stubs compile against vk_ffx.h;
up_defer_check passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
runtime/vk_up_defer.h holds, per owner, a DLSS feature, an FSR context or an image frame generation can still read,
each with the timeline value of the last submission that named it and an optional present count. upd_take releases,
in the order they were put, the entries whose value the GPU has reached and whose present count has passed; nothing
in it waits or locks, so the owner checks at its own calls on the thread that makes its other upscaler calls. It is
the piece the next commits use to take the upscalers' device-idle waits off the replaying thread: those waits stall
the present's flip, the completion thread and upload submits with the queue lock held, at every pause-screen size,
reload, window size or mode change. Nothing uses it yet.
tests/up_defer_check.py builds tests/up_defer_check.c with the header as C and as C++ (vk_ffx.cpp and vk_ffx4.cpp
include it) with -Wall -Wextra -Werror and checks both gates, the order, a full list, a null handle and a release that
puts. It passed, and failed with each of three faults put into a copy of the header.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The review found two windows in memory.c lazy_fault, where a fault holding frz_lock and lz_lock finds its address
neither armed nor in flight. It returned 0 there, and access_fault then asked for the page's state without a lock.
A write whose range a flight gave back while the write waited for frz_lock went on to thaw_fault, whose thaw released
every armed range of the 2 MiB region (lazy_drop_region_locked), each fetched under the lock. And after a release
that sets no hold (mem_lazy_skip's give-back of another buffer's range), the completion thread could arm the page
again between that 0 and access_fault's query, so a read saw an inaccessible page and went to the crash handler; that
one predates the ranges in flight.
Now a fault that finds a flight there waits for it as before. Otherwise it releases lz_lock, asks for the page's state
while it still holds frz_lock, under which no arm runs, and runs the access again when the page allows it: a read on
a readable page, a write on a read-write one. A write to a frozen page still goes on to thaw_fault, and an address
that is not ours to the crash handler.
The new late_write scenario of tests/lazy_fetch_check.py has a thread write a range whose release starts and ends
while the write waits for frz_lock: the write runs again and the region's other range stays armed. With the old
return of 0 it fails, the other range released. The read's window has no check: the arm would have to land between
two instructions. Not reproduced in the game before the change; lazy_fetch_check passed in all its scenarios.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With the setting at 0, frz_lock was held across every fault's and release's fetch again and mem_thaw released under the
lock, but two parts of the new code stayed. vk_memory.c mirror_evict and hot_check still fetched their victims'
armed ranges before taking pool_lock again, and a release still put up to 16 copies in one submission. Those are the
parts that run only with a device and have no check without one, so a device loss caused by them was the one case the
switch could not take back. Now vk_memory.c reads the setting too (evict_unlocked, once on the replay thread, where
both run): at 0 neither drops pool_lock, and the release under it fetches everything, victim by victim, as before. And
memory.c lazy_take_batch_locked takes one segment at a time at 0, so each is fetched on its own.
What stays at 0 is the table of ranges in flight: a touch of a range being fetched waits for that fetch and runs
again (a write no longer goes on to thaw_fault and releases its region), the hold is set as the range goes out, and
mem_lazy_touches counts the range. memory.c "Ranges in flight", nx2navi.ini.example and docs/building.md say so.
tests/lazy_fetch_check.py gains locked_release, the release scenario at 0: three fetches of one copy each, where the
batched release makes two of up to two copies. It fails with the cap taken out. The eviction's side has a compile
check only. Not reproduced before the change; lazy_fetch_check passed in all its scenarios.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
memory.c mem_thaw calls lazy_release_regions before it takes frz_lock, so the copy a file read's thaw needs (the
buffer's armed ranges with stale bytes) no longer holds up the other threads' faults, freezes, thaws and arms. All 20
scenarios of tests/lazy_fetch_check.py passed with that call taken out: the thaw scenario reaches the same end state
through thaw_begin_locked's lazy_drop_region_locked under the lock. The new thaw_unlocked scenario has a thread thaw a
page of a range armed with stale bytes while the fetch takes 300 ms, and freezes another region meanwhile: the freeze
returns at once, and took 300 ms with the call taken out. locked_thaw runs it with NX2NAVI_WB_LAZY_UNLOCKED=0, where
the freeze waits for the fetch as in 0.10.1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
memory.c mem_lazy_arm marks the region it protects in lz_arm_r, so that lazy_flags_locked keeps lzr set when a
release's flight in the same region ends during the arm's clean test. Without that mark the flight's end cleared lzr,
lazy_fault let a read of the arming range go, and access_fault passed it to the crash handler. All 20 scenarios of
tests/lazy_fetch_check.py passed with the mark taken out. The new arm_flight scenario has a release fetch one range for
100 ms, an arm of another range of the region wait 400 ms in its clean test, and a read of that range fault after the
flight ended: it passes, and exits with an access violation when lazy_flags_locked ignores lz_arm_r.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A CPU touch of an armed range ran its fetch (vk_memory.c lazy_fetch: a GPU copy after everything handed over, and a wait of up to 10 s) inside the fault handler under frz_lock. Every other fault, thaw, freeze and arm waited for the whole copy meanwhile: the replay's copies of guest memory (mem_copy_readable, vk_memory.c mirror_capture) and its freeze pass, the completion thread's arms, file reads and pool finalizes. A write that lost the race to the fetch went to thaw_fault after it and released every other armed range of its 2 MiB region, a second fetch. mirror_evict and hot_check fetched their victims' ranges under the exclusive pool_lock, which the completion thread's arms and every pool lookup wait for. This is AUDIT F-4-05 (C-12); the design's probe of the unchanged code had an unrelated mem_copy_readable, arm and freeze each wait 299 ms behind a 300 ms fetch, and the losing write 378.8 ms with two fetches.
A range now leaves the armed list into a table of ranges in flight (memory.c "Ranges in flight", lz_fly) in the same lz_lock hold, and stays there, its pages inaccessible, until its bytes are down and its pages read-write again (lazy_fly_run), without frz_lock. A fault on a range in flight waits for that entry alone on lz_lock's condition and runs the access again, the losing write included; a thaw, a mapping change, a release of its buffer, an arm, a freeze and mem_lazy_touches see it. Every store to lzr is made under lz_lock from the list, the entries and the region an arm is protecting (lz_arm_r, which replaces the previous commit's restore). An entry starts under frz_lock and lz_lock and ends under lz_lock alone, so a holder of frz_lock may wait for one: its fetch takes only the queue lock, which no frz_lock taker holds. mem_thaw releases what is armed in its regions before it takes frz_lock (lazy_release_regions), so file reads and pool finalizes fetch without it; a mapping change's thaw still fetches under it, as its page calls follow. The holds a fault sets (lz_hot, lz_hots) are now set as the range goes into flight, so an arm meanwhile finds them. The fetch takes several copies in one submission and one wait (MemLazyCopy), and the back end has four slots of command pool, command buffer and fence, so fetches of different ranges run at once; a release takes up to 16 armed ranges out together.
mirror_evict and hot_check now fetch what only their victims' mirrors hold before taking pool_lock exclusively, and then, under the lock, release only what was armed since (usually nothing) before the mirror leaves the list and the device address table names the imported buffer again. The design moved the release after the lock instead, which would have had the table change before the fetch: a shader of a submission still running reads through the table as it changes, and the fetch waits for every submission handed over. A victim that vk_pool_fini took off the list meanwhile is left to it; hot_check copies its log line's fields under the lock, as its records may be freed once the lock drops.
The new setting wb_lazy_unlocked (NX2NAVI_WB_LAZY_UNLOCKED, default 1, restart) set to 0 keeps frz_lock from the take-out to the end of each flight and leaves mem_thaw's releases to the thaw, as in 0.10.1; the table, the slots and the eviction order stay. The "mem: lazy write-back" line counts segments fetched; a second line gives the touches that waited for a range in flight, the longest such wait and the most ranges in flight at once; "mem: sizes" counts fetches that found every slot busy. The frz_lock and lz_lock comments say what holds what now (B-12). mem_lazy_touches has callers outside memory.c (camera_speed.c, nvn_api_cmd.c ctl_fingerprint), which now skip a range in flight instead of faulting on it and waiting.
Not reproduced in the game before the change. tests\lazy_fetch_check.py now runs 20 scenarios as their own processes: arm races, no stall of freezes, copies, arms, thaws and releases of other ranges during a fetch, joins with frz_lock held elsewhere, a losing write that keeps the region's other range armed, two fetches at once, a thaw, a mapping change and a release meeting a range in flight, an arm refused over one, the foreign give-back, the chunk mode, the old wait with wb_lazy_unlocked = 0 (joins and losing writes as well), the report lines, 80 touching threads past the 64 entries, and a freeze refused; all pass, in six runs. Taking the join before frz_lock out fails the join scenario. cfg_rows_check reports 0 problems; every runtime file compiles, and memory.c and vk_memory.c add no -Wall -Wextra warning; the test matrix passes 75 of 77, installer_test and worktree_test failing for the environment as before. vk_memory.c's slots and the eviction's order have no check without a device.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A fetch of lazily written pages runs inside the fault handler under frz_lock (memory.c lazy_gone_locked and lazy_piece_fetch_locked call vk_memory.c lazy_fetch, a GPU copy after everything handed over and a wait for it), so every other fault, thaw, freeze and arm waits for the whole copy. The line gave only the fetches' sum: the 0.10.0 session's 848 fetches took 415.8 ms in all, and how long one took, or how long another thread waited behind it, was not known. Both fetch sites now keep the longest fetch, and lazy_fault keeps the longest time a fault waited to take frz_lock; the report line prints them as "the longest X ms, faults waited up to Y ms for the freeze lock" and starts both over. No tool parses the line.
tests\lazy_fetch_check.py's new report scenario fetches for 30 and 120 ms and has a third fault wait 60 ms for frz_lock: the line reads 120.29 ms and 60.03 ms, and the next one 0.00 for both. arm_race, arm_race_slow, wb_lazy_check and heap_lazy_copy_check passed; memory.c compiles without a warning under -Wall -Wextra.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
mem_lazy_arm made a range's pages inaccessible (lazy_protect_locked: the protect, then the back end's clean test) and set the region's lzr mark only after the segment was in the list. lazy_fault returns 0 at once for an unmarked region, without taking frz_lock, so a CPU read in that window found the page inaccessible in access_fault and went to the crash handler as an access violation. A write survived only because thaw_fault waits for frz_lock. Arms come from the completion thread (vk_memory.c vk_gpu_written_back) when a submission's fence completes, which is when the game's own fence wait returns and it reads the results; the window lasts as long as the protect and the clean test, which can take tens of milliseconds while the driver pins memory.
The mark is now set before the protect, after the cheap refusals, and put back when the protect or the clean test refuses the arm. A fault from the protect on finds the mark, waits for frz_lock and then finds the segment, or the pages given back. Every store to lzr is made under frz_lock, which the arm holds, so putting it back is exact.
Not reproduced in the game before the change. The new tests\lazy_fetch_check.py reproduces it with the real Windows fault handler: a thread reads a page while it is armed and given back 20,000 times (arm_race: an access violation in 5 of 6 runs before the change), and once with a clean test of 100 ms (arm_race_slow: an access violation in every run). Both pass in every run since. wb_lazy_check, heap_lazy_copy_check, freeze_hold_check and mem_growth_check passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Whether the 64-bit generations hold past 2^32 in the game, with every texture check, mirror sync and shadow copy of a
real scene, could not be seen in play: reaching 2^32 takes 15 to 66 hours. gen_wrap_check shows it for the generation
code alone. NX2NAVI_GEN_SEED=<n> (nx2navi.ini gen_seed, diagnostics, default 0) makes the run's first generation
n + 1: with 4294966000 a run passes 2^32 within its first seconds, and with 1099511627000 the low 32 bits pass
through 0 again.
The first gen_next, before anything is stamped, reads the value (gen_seed). It is parsed with strtoull, decimal or
0x hexadecimal: strtoul stops at 2^32 - 1 where unsigned long is 32 bits, as on Windows, and would turn a seed past
32 bits into 4294967295 without a word. It is capped at 2^62, far from the 64-bit end, and only ever raises the
counter, since a lower value would leave stamps newer than it that read as written at every check. A value that is
no number, or negative, is logged and ignored; a seed in use is logged as "mem: page generations start at <n>
(NX2NAVI_GEN_SEED, a test switch)". Without the switch the only cost is one compare a generation that is true once.
cfg.c gets its metas row (0, 0..4611686018427387904, read at start), its known[] name and its sections[] entry, each
on a line of its own next to the other memory test switches (STAGING_FAIL_AT, VRAM_BALLAST_MB); nx2navi.ini.example
gets a commented gen_seed entry under [diagnostics]. The settings menu does not show it.
tests/gen_wrap_check runs the generation code a second way: the counter left at 0 and the seed taken from
NX2NAVI_GEN_SEED, for 4294967288, 0xfffffffff8, 0x7ffffffffffffff8 (capped), 12abc, -8, 0 and unset, each with the
start and the log line expected of it, then the same compares across the boundary. With strtoul in place of strtoull
two of those runs fail. Checks that passed: gen_wrap_check, gen_width_check, cfg_rows_check (0 problems),
pass_cc_check on vk_memory.c and cfg.c, no new -Wall -Wextra warnings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
vk_memory.c handed out page write generations from a 32-bit counter (gen_now) and every consumer compared them as
32-bit values, with 0 meaning "never stamped". At 124 to 336 generations a frame (a 0.10.0 log with trace = perf) the
counter wrapped after about 66 h of play at 60 fps, 27.6 h at 144 and 15 to 17 h at 240. Past the wrap a stamp made
before it compared newer than every later check and one made after it older: textures and VRAM mirrors missed the
CPU's writes (stale textures, and the lazy fetch copying older mirror bytes over a CPU write), pages nothing wrote read
as written at every check (a hash or upload every time, and every such page of a mirrored pool copied up at each
sync), and the generation handed out at the wrap was 0. At 64 bits nothing wraps: 2^64 lasts millions of years at
336 a frame and 240 fps. Below 2^32 every value and every compare is the same as before.
What changed: gen_now, both tables (gen_chunk, cgen_chunk), reg_max and creg_max, PoolRec's synced_gen, mgen and sgen,
dirty_gen and DIRTY_UNSYNCED, VSample.mgen, every local and parameter that holds a generation in vk_memory.c,
vk_texture.c (Tex.wgen and gpu_gen, PEnt.since, storage_rewritten and its callers, sweep_hard's write-back check
included) and tex_probe.c, and vk.h's vk_mem_changed, vk_mem_changed_snap and vk_mem_unchanged(_why)_locked. The verify
mode's mismatch line prints the generations with %llu. The readers that take no gen_lock (lazy_clean, cpu_wrote_since,
verify_cb, vk_mem_sizes) stay lock-free at 64 bits: the slots, mgen, gen_now and a new chunk pointer are now stored
atomically (relaxed, or release where a reader acquires), the same instructions on x86-64, and a static assertion
holds that 64-bit atomics are lock-free. No lock, wait or thread changes.
gen_next, which hands out every generation, no longer warns at 0xC0000000 and up or at the wrap, which cannot come;
it logs one line when the count reaches 2^32, where earlier versions ran out, with the frame, the hours and the rate.
Cost: the tables take 8 bytes a page instead of 4 (512 KiB more per 256 MiB of guest address space stamped, both
tables together, allocated as before when a chunk is first stamped), reg_max and creg_max 2 MiB each instead of 1,
Tex and PoolRec 8 bytes more, PEnt 8 more. The page walks read 8 bytes a slot; a micro-benchmark of the walk measured no
difference.
Not reproduced in the game before the change: the wrong answers were reproduced by compiling vk_memory.c's generation
code with the counter seeded below 2^32. tests/gen_wrap_check now runs that code, with the readers outside its
section, across 2^32, 2^40 and 2^63 and checks that the compares answer as below the boundary; on the code before the
change 19 of its checks fail at 2^32 and the other two seeds do not fit the counter. The new tests/gen_width_check
compiles vk_memory.c, vk_texture.c and tex_probe.c with -Wshorten-64-to-32, -Wformat and -Wincompatible-pointer-types
as errors, so a 32-bit local left behind is found (shown by reverting single locals, fields and formats). Checks that
passed: gen_wrap_check, gen_width_check, pass_cc_check of every runtime file, no new -Wall -Wextra warnings in the
three files, tex_release_check, tex_compact_check, wb_lazy_check, gpa_check, mem_growth_check, mem_budget_check.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The review found that nothing on the release path ran the new export check: tools/make_launcher.py build_app and
tools/make_kit.py copied whatever --dlssg DLL they were given, and the release script still named the DLL of 0.8.0,
which lacks nx2sl_reflex2. Shipped that way, Reflex would never hold the frame limit while players still got this
branch's other changes. make_launcher.py check_dlssg and make_kit.py main now run build_dlssg.check on the --dlssg DLL
beside the translator's check, before anything is resolved, downloaded or deleted; build_dlssg.check also stops on a
missing file and on a file that is no DLL, with a message instead of a traceback. tests/dlssg_exports_check.py adds a
DLL with every declared export (it passes), the two-export DLL, a missing file and two files that are no DLL (they
stop, naming why), make_launcher.check_dlssg on both DLLs, and that both scripts' main makes the check before
streamline_pin.resolve and shutil.rmtree; it fails when the call is taken out of make_launcher.py main. The DLL built
from this branch passes make_launcher.check_dlssg; the DLL of 0.8.0 is refused, naming nx2sl_reflex2.
docs/launcher.md and docs/building.md "Checks" say so.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The runtime takes a missing optional export (nx2sl_reflex2, nx2sl_frames_max) for an older DLL and leaves the
feature out with one log line, so a release that ships a DLL built from older sources than its runtime loses Reflex's
frame limit without an error; the release script took the DLL of 0.8.0 from a fixed path. After the build,
build_dlssg.py now reads every PFN_nx2sl_<name> in runtime/dlssg/nx2_sl.h and the DLL's export directory (a small
PE reader of its own, PE32 and PE32+) and exits naming each export that is missing; --check <dll> makes the same
check without building. tests/dlssg_exports_check.py builds a DLL with two of three declared exports with MSYS2's
clang and checks that the third is named, and that nx2_sl.h's own declarations are read in full; docs/building.md
"Checks" and docs/launcher.md name both. The DLL built from this branch with the pinned Streamline 2.14.16 SDK passes
(15 exports); the one built before nx2sl_reflex2 fails, naming it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
